UNDERSTANDING CUSTOMERS - PROFILING AND

SEGMENTATION
Mircea Andrei SCRIDON
Babes Bolyai University of Cluj-Napoca

Abstract:
In any industry, the first step to finding and creating profitable customers is
determining what drives profitability. This leads to better prospecting and
more successful customer relationship management. Any company can
segment and profile their customer base to uncover those profit drivers using
the knowledge of their customers, products, and markets. Or they can use
data-driven techniques to find natural clusters in their customer or prospect
base. Whatever the method, the process will lead to knowledge and
understanding that is critical to maintaining a competitive edge.

Key words: profiling, segmentation, penetration analysis, cluster analysis.

The methodologies discussed in out offers for products and services—

this paper describe several techniques without a clue of what their best
and applications for understanding customer looks like. For every company
customers. Common sense tells us that in every industry, this is the most
it is a good first step to successful important first step to profitable
customer relationship management. It is marketing. Similar to modeling, before a
also an important step for effective company begins any profiling or
prospecting. In other words, once segmentation project, it is important to
companies know what customer establish their objective. This is crucial
attributes and behaviors are currently because it will affect the way in which to
driving their profitability, they can use approach the task. The objective can be
these to direct their prospecting efforts explained by reviewing the definitions of
as well. The first step in effective profiling and segmentation.
prospecting is learning how to find Profiling is exactly what it implies:
prospects that look like customers for the act of using data to describe or
the company. It is also useful to profile a group of customers or
segment and profile the prospect base prospects. It can be performed on an
to assist acquisition efforts. The goal in entire database or distinct sections of
both cases is to identify what drives the database. The distinct sections are
customer profitability. known as segments. Typically they are
The second half of the paper mutually exclusive, which means no one
details the process using a case study. can be a member of more than one
The study is from the household segment.
industry, in which it will show some Segmentation is the act of splitting
profile and penetration analyses. Next, it a database into distinct sections or
will illustrate the use of cluster analysis segments. There are two basic
to discover segments. approaches to segmentation: market
driven and data driven. Market-driven
What is the importance of approaches allow managers to use
understanding customers? characteristics that they determine to be
Studies show that many important drivers of their business. In
companies operate for years— pumping other words, they preselect the
characteristics that define the recently purchased something from a
segments. This is why defining the company, it is more likely to make
objective is so critical. The ultimate another purchase than someone who
plans for using the segments will did not recently make a purchase.
determine the best method for creating Frequency. This value is the
them. On the other hand, data-driven number of purchases. It can be the total
approaches use techniques such as of purchases within a specific time
cluster analysis or factor analysis to find frame or include all purchases. This
homogenous groups. This might be characteristic is second to recency in
useful if companies are working with predictive power for response. Again, it
data about which they have little is quite intuitive as to why it relates to
knowledge. future purchases.
Monetary value. This value is the
Types of Profiling and total currency amount. Similar to
Segmentation frequency, it can be within a specific
If a company have never done any time frame or include all purchases. Of
segmentation or modeling, its customer the three, this characteristic is the least
base may seem like a big blob that powerful when it comes to predicting
behaves a certain way, depending on response. But when used in
the latest stimulus. If the company does combination, it can add another
a little digging, it will find a variety of dimension of understanding.
demographic and psychographic These three characteristics can be
characteristics as well as a multitude of used alone or in combination with other
buying behaviors, risk patterns, and characteristics to assist in CRM efforts.
levels of profitability among the
members of its database. This is the Demographic
beauty of segmentation and profiling. The emphasis is on finding the
Once companies understand the distinct individual who may not fit the local
groups within the database, they can demographic profile. In reality, though,
use this knowledge for product many people who live in the same area
development, customer service behave in a similar fashion. There are
customization, media and channel many sources of demographic data.
selection, and targeting selection. Many sources are collected at the
individual level with enhancements from
RFM: Recency, Frequency, the demographics of the surrounding
Monetary Value geographic area. Segmenting by values
One of the most common types of such as age, gender, income, and
profiling originated in the catalog marital status can assist in product
industry. Commonly called RFM, it is a development, creative design, and
method of segmenting customers on targeting. There are several methods for
their buying behavior. Its use is primarily using demographics to segment a
for improving the efficiency of marketing database and/or build customer profiles.
efforts to existing customers. It is a very
powerful tool that involves little more Life Stage
than creating segments from the three With few exceptions, our lives
groups. follow patterns that change over time to
Recency. This value is the number meet our needs. These patterns are
of months since the last purchase. It is clustered into groups defined by
typically the most powerful of the three demographics like age, gender, marital
characteristics for predicting response status, and presence of children to form
to a subsequent offer. This seems quite life stage segments. Life stage
logical. It says that if a consumer has segments are typically broken into
176
young singles; couples or families; perspective throughout the
middle-aged singles, couples, or development, testing, and analysis
families; and older singles or couples. stages, companies are more assured of
Additional enhancements can be meeting their goals, maximizing their
achieved by overlaying financial, profitability and improving their
behavioral, and psychographic data to customers' long-term behavior.
create well-defined homogeneous
segments. Understanding these Profiling and penetration
segments provides opportunities for analysis of a catalog company's
businesses to develop relevant products customers
and fine-tune their marketing strategies. The following case study is about a
The market segments should pass company which offers household items.
the following rules in order to be taken It has been operating for six years and
into consideration: now has a database of 14244
• They should be relevant to the customers. It is interested in expanding
business objective its customer base. It is therefore looking
• They should be understandable for ways to identify good prospects from
and easy to characterize outside sources. The first step is to
• They should be large enough to perform RFM analysis.
warrant a special offering
• They should be easy to develop RFM Analysis
unique offerings As it has been mentioned earlier,
After testing which of the segments recency, frequency, and monetary value
have passed the above rules, a are typically the strongest drivers of
company should evaluate the results response for a catalog company. To
behaviorally and financially to determine discover the effects of these measures
which segmentations and offerings on company's database, the relevant
should be expanded to the target variables are identified in the database:
population, and also how good did they last purchase (lp) - Months since
perform against the business objectives, last purchase or recency.
set by it. Managers often fail to number of purchases (np) -
remember the business objectives until Number of purchases in the last 36
it is too late. It is critical that they have months or frequency.
designed the segmentations to satisfy a total purchases (tp) - Total
business objective and that they have monetary amount of purchases in the
evaluated the market tests based on last 36 months or monetary value.
those same business objectives. If not, The first step is to get a distribution
the fall-out could be costly on other of the customers' general patterns. The
fronts, such as lower profitability, tables below provide a good overview of
reduced product usage, or negative customer buying habits for the
changes in attitude or expectations. By company:
keeping the business objectives in

Table 1
Recency of purchases
Lp (months) Percent Cumulative percent
0-2 17,1 17,1
3-5 13,6 30,7
6-8 41,2 71,9
9-12 28,1 100,0

177
 Table 2
Number of purchases
Np Percent Cumulative percent
0-1 30 30
2-6 43,1 73,1
7-14 19,8 92,9
15-20 7,1 100,0

Table 3
Total purchases
Tp (€) Percent Cumulative percent
<100 37,9 37,9
100-200 24 61,9
200-300 10,2 72,1
300-400 5,4 77,5
400-500 4,1 81,6
500+ 18,4 100,0

It can be observed that the activity to see how these three drivers
majority of customers haven't affect response. The database is sorted
purchased anything for at least six by recency and created quintiles from it
months. A large percentage of (equal fifths of the file). By calculating
customers made between two and six the response rate for each quintile, the
purchases in the last year with 92,9% relationship between recency and
making fewer than seven purchases. response can be determined.
The total monetary value of yearly total It can be seen that the measure
purchases is mainly below 100 €, with with the strongest relationship to
77,5% below 400 €. response is recency according to table
The next step is to look at the 4.
response rate from a recent mailing

178
 Table 4 Table 5 Table 6

Quantile – Recency Quantile – Frequency Quantile – Monetary

Response Response
Recency Response Monetary
rate Frequency rate
rate
Quantile Quantile
Quantile
Q1 0,21 Q1 0,078
Q1 0,145
Q2 0,082 Q2 0,041
Q2 0,03
Q3 0,018 Q3 0,037
Q3 0,019
Q4 0,012 Q4 0,012 Q4 0,022

Q5 0,005 Q5 0,011 Q5 0,015

The following figure compares recency, frequency, and monetary value as

they relate to response.

Figure 1. Quantile - Response rate

Again, it can be seen that the an effective method for comparing the
recency of purchase is the strongest distribution of the customer base to the
driver. This is a valuable piece of general population. Many companies
information and can be used to target sell lists that cover a broad base of the
the next campaign. In fact, many population. The methodology is simple.
companies include a new catalog in First it procedure begins with a
every order. This is a very inexpensive frequency distribution of some basic
way to take advantage of recent demographic variables.
purchase activity. In this case, the variables age and
income are selected. The figure below
Penetration Analysis shows the output from for these two
As it has been mentionated earlier, variables. This gives information about
the study company wants to explore the distribution of the company’s
cost-effective techniques for acquiring customers. It can be noticed how 35%
new customers. Penetration analysis is
179
of the customers are between the ages of 41 and 44.

Table 7
Age – Customers
Age Percent Cumulative percent
<25 0,3 0,3
26-30 1,7 2
31-35 10,2 12,2
36-40 28,4 40,6
41-44 35 75,6
45-54 15,3 90,9
55-64 7,6 98,5
65+ 1,5 100

Table 8
Income – Customers
Income (€) Percent Cumulative percent
<400 1,1 1,1
400-700 26,4 27,5
700-900 27,2 54,7
900-1200 25,5 80,2
1200-1500 18,2 98,4
1500+ 1,6 100

In order to make use of this finding to the general population. The

information for new acquisition next table showes similar profiles for the
marketing, it is needed to compare this general population:

Table 9
Age – General population
Age Percent Cumulative percent
<25 1,4 1,4
26-30 2,3 3,7
31-35 12 15,7
36-40 26,9 42,6
41-44 30,5 73,1
45-54 18,4 91,5
55-64 7,6 99,1
65+ 0,9 100

Table 10
Income – General population
Income (€) Percent Cumulative percent
<400 1,6 1,6
400-700 25,2 26,8
700-900 28,4 55,2
900-1200 23,9 79,1
1200-1500 16,1 95,2
1500+ 4,8 100

180
 It can be noticed how the distributions as the first table except this
second figure displays the same time they are on the general population.

Figure 2. Age distribution among customers and general population

The figure above shows a market together and creates a measure called
comparison graph of age. It provides a a penetration index. This is derived by
graphical display of the differences in dividing the customer percentage by the
distribution for the various age market percentage for each group and
groupings. multiplying by 100.
The following table brings the
information from the two analyses

Table 11
Penetration analysis – Age
Percent of Percent of Penetration
Age
customers market index
<25 0,3 1,4 21,42 %
26-
1,7 2,3 73,91 %
30
31-
10,2 12 85 %
35
36-
28,4 26,9 105,57 %
40
41-
35 30,5 114,75 %
44
45-
15,3 18,4 83,15 %
54
55-
7,6 7,6 100 %
64

181
 Table 12
Penetration analysis – Income
Percent of Percent of Penetration
Income
customers market index
<400 1,1 1,6 68,75 %
400-700 26,4 25,2 104,76 %
700-900 27,2 28,4 95,77 %
900-1200 25,5 23,9 152 %
1200-
18,2 16,1 106,69 %
1500
1500+ 1,6 4,8 33,33 %

The company would be wise to by randomly assigning cluster seeds or

see new customers in the 36–44 age centers. The number of seeds is equal
group. This age range is more to the number of clusters requested.
prominent in its customer base than in Each observation is assigned to the
the general population. Also wise would nearest seed. The seed is then
be to take in consideration new reassigned to the mean in each cluster.
customers whose income is situated in The process is repeated until the
the 400-700 € group and 1200-1500 € change in the seed becomes sufficiently
group. small.
To illustrate the methodology, two
Performing cluster analysis to variables (age and income) are used
discover customer segments from the company data in this case
Cluster analysis is a family of study. Before the cluster analysis
mathematical and statistical techniques begins, the variables must be
that divides data into groups with similar standardized. Because the clustering
characteristics. Clustering uses algorithm used depends on distance
Euclidean distance to group between variable values, the scales of
observations together that are similar the variables must be similar.
across several characteristics, while Otherwise, the variable with the largest
attempting to separate observations that scale will dominate the clustering
are dissimilar across those same procedure.
characteristics. It is a process with In the next figure, the output
many opportunities for guidance and displays the distance from the seeds to
interpretation. the farthest point as well as the
Several algorithms are used in distances between clusters. The cluster
clustering. In this case study, a means do show a notable difference in
clustering algorithm is used that is values for age and income.
designed for large data sets. It begins

182
 Table 13
Cluster summary
Maximum distance
Distance
RMS Std. from seed to Nearest
Cluster Frequency between cluster
deviation observation cluster
centroids
1 10185 0,71 3,21 2 1,62
2 3512 0,82 3,56 1 1,62
3 547 0,98 4,02 1 3,73

Statistics for variables

Variable Total STD Within STD R-squared
Age 1 0,58 0,52
Income 1 0,69 0,47
Over-all 1 0,75 0,51

Pseudo F statisitc = 24321,25

Approximate expexted over-all R-squared = 0,586

Cluster means
Cluster Age Income
1 0,54247 0,10581
2 1,31591 -0,42131
3 -0,21656 2,98939

The company can now tailor their Data must be relevant to the business
marketing campaigns to each group objectives. The process involves
separately. Similar to the profile reviewing all data to determine only the
analysis, understanding the segments necessary elements because collecting
can improve targeting and provide and analyzing data on all customers or
insights for marketers to create relevant prospects is very time-consuming and
offers. expensive.
The segmentation process means
Conclusions selecting a method that is appropriate
A sucesful profiling and for the situation. There are three
segmentation process demands that a segmentation methods that could be
company should define its business employed: predefined segmentation,
objectives. At the start of any statistical segmentation, or hybrid
segmentation process, management segmentation.
should agree on and clearly state their The predefined segmentation
goals using language that reflects method allows the analyst to create the
targeting and measurement. Business segment definitions based on prior
objectives can be (1) new account, experience and analysis. In this case,
sales, or usage driven; (2) new product the data is known, the work involves a
driven; (3) profitability driven; or (4) limited number of variables, and a
product or service positioning driven. limited number of segments are
Furthermore types of data could determined. The appropriate segments
include survey, geo-demographic will be defined and selected based on
overlays, and transactional behavior.
183
the business objective and the
knowledge of the customer base. Hybrid segmentation allows the
The statistical method should be analyst to combine predefined
employed when there are many segmentation with statistical
segments involved and there is little or segmentation, in any order, based on
no experience with the population being the success in deriving segments.
investigated.

REFERENCES

[1] Buiga, Anuţa, Florea I., Lazăr, Dorina, Parpucea, Ilie, Statistică inferenţială,
Presa Universitară Clujeană, Cluj-Napoca, 2000.

[2] Hand, David, Mannila, Heikki, Smyth, Padhraic, Principles of Data Mining, The
MIT Press, Cambridge, MA, 2001.

[3] Hughes, Arthur M, Strategic Database Mining, Probus Publishing, Chicago,

1994.

[4] Pop, Marcel C., Implicaţiile pieţei asupra planificării strategice de marketing,
Risoprint, Cluj-Napoca, 2005.

[5] Rud, Olivia Parr, Data Mining Cookbook: Modeling Data for Marketing, Risk, and
Customer Relationship Management, John Wiley & Sons, New York, 2001.

184